My recent posts have explored how publishers are working with social platforms to expand audience and adapt story telling formats (see Publishers & Platforms In a Relationship, and Platforms as Publishers: 6 Key Takeaways for Brands). They reported the experiences of social teams and editors at some of the largest broadcast, print daily and native web outlets.
Those featured, however, didn’t go into detail on the role of advertising to boost reach.
At last week’s NY Data Science Meetup (at Metis NYC) we learned how the Huffington Post, the largest social publisher, is using data science to better understand which articles can benefit from a promotional push. Their efforts have propelled merely popular stories into through-the-roof viral successes.
The meetup was about Data Science in the Newsroom. Geetu Ambwani, Principal Data Scientist at Huffington Post, recalled the days when their editors monitored searches trending on Google to inform content creation and curation. Since then it is a new game, as more people are discovering and consuming news through social media.
In an age of distributed news, HuffPo needed a new approach.
Data across the Content Life Cycle
Geetu discussed the role of data in the content life cycle spanning creation, distribution and consumption. For creation, there are tools to discover trends, enhance and optimize content, and flag sensitive topics. Their RobinHood platform improves image usage and the all-important headline.
Geetu’s favorite part, she said, was exploring the “content gap” between what they write and what people want to read. It’s a tension that must be carefully considered – otherwise writers might be tempted to focus on fluff pieces vs. important news stories.
When it comes to consumption, data can be used to improve the user experience – e.g. via recommendations and personalization.
Project Fortune Teller: Data Predict Viral Success
Geetu and her team turned to data science to help with distribution. “The social networks are the new home page – we need to be where the audience is,” she said.
Only a small percentage of their stories get significant page views on the web. Performance on social often varies by platform. The team honed the content mix for each to improve engagement. Part of this was determining which articles out of the 1000 daily stories should get an extra boost.
Geetu wondered if they could mine data to spot the ones that have “legs” beyond early popularity. With this info in hand, they could promote these with high value ads, and populate Trending Now and Recommendation widgets to further boost sharing and reach.
And thus , Project Fortune Teller was born. The team looked for winners according to a range of data such as web traffic growth, and social consumption and sharing. But it was no easy task. There are many variables to consider. They needed to determine the optimal time window, as some articles take a bit longer to start to trend. Finally, they intentionally excluded hot news stories, instead focusing on evergreen content that was resonating.
Geetu and her team mined historical data, using time series analysis to build a model (for more details, see this SlideShare presentation). They notified the content promotion staff when there was a likely winner. The resulting quick action turned popular articles into viral successes.
The conclusion? Machine learning is a key driver of success for predicted content.
Share article on